Source code for mrmustard.physics.gaussian_integrals

# Copyright 2024 Xanadu Quantum Technologies Inc.

# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at

#     http://www.apache.org/licenses/LICENSE-2.0

# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""
This module contains gaussian integral functions and related helper functions.
"""

from typing import Sequence, Tuple
import numpy as np
from mrmustard import math
from mrmustard.utils.typing import ComplexMatrix, ComplexVector


[docs] def real_gaussian_integral( Abc: tuple, idx: tuple[int, ...], ): r"""Computes the Gaussian integral of the exponential of a real quadratic form. The integral is defined as (note that in general we integrate over a subset of `m` dimensions): .. :math:: \int_{R^m} F(x) dx, where :math:`F(x) = \textrm{exp}(0.5 x^T A x + b^T x)` Here, ``x`` is an ``n``-dim real vector, ``A`` is an ``n x n`` real matrix, ``b`` is an ``n``-dim real vector, ``c`` is a real scalar. The integral indices are specified by ``idx``. Arguments: Abc: the ``(A,b,c)`` triple idx: the tuple of indices of the x variables Returns: The ``(A,b,c)`` triple of the result of the integral. """ A, b, c = Abc if not idx: return A, b, c not_idx = tuple(i for i in range(A.shape[-1]) if i not in idx) M = math.gather(math.gather(A, idx, axis=-1), idx, axis=-2) bM = math.gather(b, idx, axis=-1) if math.asnumpy(not_idx).shape != (0,): D = math.gather(math.gather(A, idx, axis=-1), not_idx, axis=-2) R = math.gather(math.gather(A, not_idx, axis=-1), not_idx, axis=-2) bR = math.gather(b, not_idx, axis=-1) T = math.transpose L = T(math.solve(T(M), T(D))) A_post = R - math.matmul(L, T(D)) b_post = bR - math.matvec(L, bM) else: A_post = math.astensor([]) b_post = math.astensor([]) c_post = ( c * math.sqrt((2 * np.pi) ** len(idx), math.complex128) * math.sqrt((-1) ** len(idx) / math.det(M), math.complex128) * math.exp(-0.5 * math.sum(bM * math.solve(M, bM))) ) return A_post, b_post, c_post
[docs] def complex_gaussian_integral( Abc: tuple, idx_z: tuple[int, ...], idx_zconj: tuple[int, ...], measure: float = -1 ): r"""Computes the Gaussian integral of the exponential of a complex quadratic form. The integral is defined as (note that in general we integrate over a subset of 2m dimensions): :math:`\int_{C^m} F(z) d\mu(z)` where :math:`F(z) = \textrm{exp}(-0.5 z^T A z + b^T z)` Here, ``z`` is an ``n``-dim complex vector, ``A`` is an ``n x n`` complex matrix, ``b`` is an ``n``-dim complex vector, ``c`` is a complex scalar, and :math:`d\mu(z)` is a non-holomorphic complex measure over a subset of m pairs of z,z* variables. These are specified by the indices ``idx_z`` and ``idx_zconj``. The ``measure`` parameter is the exponent of the measure: :math: `dmu(z) = \textrm{exp}(m * |z|^2) \frac{d^{2n}z}{\pi^n} = \frac{1}{\pi^n}\textrm{exp}(m * |z|^2) d\textrm{Re}(z) d\textrm{Im}(z)` Note that the indices must be a complex variable pairs with each other (idx_z, idx_zconj) to make this contraction meaningful. Please make sure the corresponding complex variable with respect to your Abc triples. For examples, if the indices of Abc denotes the variables ``(\alpha, \beta, \alpha^*, \beta^*, \gamma, \eta)``, the contraction only works with the indices between ``(\alpha, \alpha^*)`` pairs and ``(\beta, \beta^*)`` pairs. Arguments: A,b,c: the ``(A,b,c)`` triple idx_z: the tuple of indices of the z variables idx_zconj: the tuple of indices of the z* variables measure: the exponent of the measure (default is -1: Bargmann measure) Returns: The ``(A,b,c)`` triple of the result of the integral. Raises: ValueError: If ``idx_z`` and ``idx_zconj`` have different lengths. """ A, b, c = Abc if len(idx_z) != len(idx_zconj): raise ValueError( f"idx_z and idx_zconj must have the same length, got {len(idx_z)} and {len(idx_zconj)}" ) n = len(idx_z) idx = tuple(idx_z) + tuple(idx_zconj) if not idx: return A, b, c not_idx = tuple(i for i in range(A.shape[-1]) if i not in idx) I = math.eye(n, dtype=A.dtype) Z = math.zeros((n, n), dtype=A.dtype) X = math.block([[Z, I], [I, Z]]) M = math.gather(math.gather(A, idx, axis=-1), idx, axis=-2) + X * measure bM = math.gather(b, idx, axis=-1) c_post = ( c * math.sqrt((-1) ** n / math.det(M)) * math.exp(-0.5 * math.sum(bM * math.solve(M, bM))) ) if math.asnumpy(not_idx).shape != (0,): D = math.gather(math.gather(A, idx, axis=-1), not_idx, axis=-2) R = math.gather(math.gather(A, not_idx, axis=-1), not_idx, axis=-2) bR = math.gather(b, not_idx, axis=-1) A_post = R - math.matmul(D, math.inv(M), math.transpose(D)) b_post = bR - math.matvec(D, math.solve(M, bM)) else: A_post = math.zeros((0, 0), dtype=A.dtype) b_post = math.zeros((0,), dtype=b.dtype) return A_post, b_post, c_post
[docs] def join_Abc( Abc1: Tuple[ComplexMatrix, ComplexVector, complex], Abc2: Tuple[ComplexMatrix, ComplexVector, complex], ): r"""Joins two ``(A,b,c)`` triples into a single ``(A,b,c)`` triple by block addition of the ``A`` matrices and concatenating the ``b`` vectors. Arguments: Abc1: the first ``(A,b,c)`` triple Abc2: the second ``(A,b,c)`` triple Returns: The joined ``(A,b,c)`` triple """ A1, b1, c1 = Abc1 A2, b2, c2 = Abc2 A12 = math.block_diag(math.cast(A1, "complex128"), math.cast(A2, "complex128")) b12 = math.concat([b1, b2], axis=-1) c12 = math.outer(c1, c2) return A12, b12, c12
[docs] def join_Abc_real( Abc1: Tuple[ComplexMatrix, ComplexVector, complex], Abc2: Tuple[ComplexMatrix, ComplexVector, complex], idx1: Sequence[int], idx2: Sequence[int], ): r"""Direct sum of two ``(A,b,c)`` triples into a single ``(A,b,c)`` triple, where indices corresponding to the same variable are "fused together", by considering their Bargmann function has having the same variables. For example ``idx1=(0,1,2)`` and ``idx2=(1,2,3)`` means that indices 1 and 2 will be fused because they are present on both tuples. This is useful for computing real Gaussian integrals where the variable on either object is the same, rather than a pair of conjugate variables for complex Gaussian integrals. Arguments: Abc1: the first ``(A,b,c)`` triple Abc2: the second ``(A,b,c)`` triple idx1: the indices of the first ``(A,b,c)`` triple to fuse idx2: the indices of the second ``(A,b,c)`` triple to fuse Returns: The joined ``(A,b,c)`` triple with the order [idx1(or idx2), not_idx2]. """ A1, b1, c1 = Abc1 A2, b2, c2 = Abc2 if len(idx1) != len(idx2): raise ValueError( f"idx1 and idx2j must have the same length, got {len(idx1)} and {len(idx2)}" ) if (len(idx1) > A1.shape[-1]) or (len(idx2) > A2.shape[-1]): raise ValueError(f"idx1 and idx2 must be valid, got {len(idx1)} and {len(idx2)}") not_idx1 = tuple(i for i in range(A1.shape[-1]) if i not in idx1) not_idx2 = tuple(i for i in range(A2.shape[-1]) if i not in idx2) A1_idx_idx = math.gather(math.gather(A1, idx1, axis=-1), idx1, axis=-2) b1_idx = math.gather(b1, idx1, axis=-1) if not_idx1: A1_idx_notidx = math.gather(math.gather(A1, not_idx1, axis=-1), idx1, axis=-2) A1_notidx_idx = math.gather(math.gather(A1, idx1, axis=-1), not_idx1, axis=-2) A1_notidx_notidx = math.gather(math.gather(A1, not_idx1, axis=-1), not_idx1, axis=-2) b1_notidx = math.gather(b1, not_idx1, axis=-1) A2_idx_idx = math.gather(math.gather(A2, idx2, axis=-1), idx2, axis=-2) b2_idx = math.gather(b2, idx2, axis=-1) if not_idx2: A2_idx_notidx = math.gather(math.gather(A2, not_idx2, axis=-1), idx2, axis=-2) A2_notidx_idx = math.gather(math.gather(A2, idx2, axis=-1), not_idx2, axis=-2) A2_notidx_notidx = math.gather(math.gather(A2, not_idx2, axis=-1), not_idx2, axis=-2) b2_notidx = math.gather(b2, not_idx2, axis=-1) if math.asnumpy(not_idx1).shape == (0,): A12 = math.block([[A1 + A2_idx_idx, A2_notidx_idx], [A2_idx_notidx, A2_notidx_notidx]]) b12 = math.concat([b1 + b2_idx, b2_notidx], axis=-1) c12 = math.outer(c1, c2) elif math.asnumpy(not_idx2).shape == (0,): A12 = math.block([[A2 + A1_idx_idx, A1_notidx_idx], [A1_idx_notidx, A1_notidx_notidx]]) b12 = math.concat([b2 + b1_idx, b1_notidx], axis=-1) c12 = math.outer(c1, c2) else: O_n = math.zeros((len(not_idx1), len(not_idx2)), math.complex128) A12 = math.block( [ [A1_idx_idx + A2_idx_idx, A1_idx_notidx, A2_idx_notidx], [A1_notidx_idx, A1_notidx_notidx, O_n], [A2_notidx_idx, O_n.T, A2_notidx_notidx], ] ) b12 = math.concat([b1_idx + b2_idx, b1_notidx, b2_notidx], axis=-1) c12 = math.outer(c1, c2) return A12, b12, c12
[docs] def reorder_abc(Abc: tuple, order: Sequence[int]): r""" Reorders the indices of the A matrix and b vector of an (A,b,c) triple. Arguments: Abc: the ``(A,b,c)`` triple order: the new order of the indices Returns: The reordered ``(A,b,c)`` triple """ A, b, c = Abc if order: A = math.gather(math.gather(A, order, axis=-1), order, axis=-2) b = math.gather(b, order, axis=-1) if len(c.shape) == len(order): c = math.transpose(c, order) return A, b, c
[docs] def contract_two_Abc( Abc1: Tuple[ComplexMatrix, ComplexVector, complex], Abc2: Tuple[ComplexMatrix, ComplexVector, complex], idx1: Sequence[int], idx2: Sequence[int], ): r""" Returns the contraction of two ``(A,b,c)`` triples with given indices. Note that the indices must be a complex variable pairs with each other to make this contraction meaningful. Please make sure the corresponding complex variable with respect to your Abc triples. For examples, if the indices of Abc1 denotes the variables ``(\alpha, \beta)``, the indices of Abc2 denotes the variables ``(\alpha^*,\gamma)``, the contraction only works with ``idx1 = [0], idx2 = [0]``. Arguments: Abc1: the first ``(A,b,c)`` triple Abc2: the second ``(A,b,c)`` triple idx1: the indices of the first ``(A,b,c)`` triple to contract idx2: the indices of the second ``(A,b,c)`` triple to contract Returns: The contracted ``(A,b,c)`` triple """ Abc = join_Abc(Abc1, Abc2) return complex_gaussian_integral( Abc, idx1, tuple(n + Abc1[0].shape[-1] for n in idx2), measure=-1.0 )